Potential-Based Greedy Matching for Dynamic Delivery Pooling

2025-02-24

AI Summary

This paper addresses the real-time order batching problem in dynamic food-delivery scenarios, aiming to minimize total travel distance while balancing immediate matching gains against the strategic retention of orders to enhance future batching opportunities. To this end, we propose a potential-energy-driven greedy strategy (PB), which—uniquely—uses geographic potential energy of orders as the primary decision criterion, eliminating reliance on demand forecasting or historical statistics. We further design a lightweight, spatial-topology-based heuristic algorithm and provide a worst-case theoretical performance guarantee, proving its superiority over naive greedy approaches. Extensive simulations across multiple scenarios—using both synthetic data and real-world order traces from Meituan—demonstrate that PB consistently outperforms state-of-the-art batch-processing and prediction-aware baselines in distance savings, validating its effectiveness and irreplaceability for global route optimization.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Search and Optimization: Heuristic Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: The sharing economy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We study the problem of pooling together delivery orders into a single trip, a strategy widely adopted by platforms to reduce total travel distance. Similar to other dynamic matching settings, the pooling decisions involve a trade-off between immediate reward and holding jobs for potentially better opportunities in the future. In this paper, we introduce a new heuristic dubbed potential-based greedy (PB), which aims to keep longer-distance jobs in the system, as they have higher potential reward (distance savings) from being pooled with other jobs in the future. This algorithm is simple in that it depends solely on the topology of the space, and does not rely on forecasts or partial information about future demand arrivals. We prove that PB significantly improves upon a naive greedy approach in terms of worst-case performance on the line. Moreover, we conduct extensive numerical experiments using both synthetic and real-world order-level data from the Meituan platform. Our simulations show that PB consistently outperforms not only the naive greedy heuristic but a number of benchmark algorithms, including (i) batching-based heuristics that are widely used in practice, and (ii) forecast-aware heuristics that are given the correct probability distributions (in synthetic data) or a best-effort forecast (in real data). We attribute the surprising unbeatability of PB to the fact that it is specialized for rewards defined by distance saved in delivery pooling.

Problem

• Dynamic delivery order pooling
• Potential-based greedy algorithm
• Minimizing travel distance

Innovation

• Potential-Based Greedy Matching
• Topology-Dependent Algorithm
• Distance Savings Optimization